whisper-new-nnat-20h-maxcos-ecapa

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.4646
• Wer: 20.6950

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 109
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.322	0.2986	109	0.5722	26.5302
0.421	0.5973	218	0.5081	23.2661
0.3857	0.8959	327	0.4898	21.8892
0.3523	1.1945	436	0.4843	21.9360
0.3301	1.4932	545	0.4772	21.4864
0.3187	1.7918	654	0.4686	21.2336
0.328	2.0904	763	0.4704	20.9713
0.2973	2.3890	872	0.4684	20.6201
0.2979	2.6877	981	0.4659	20.5264
0.2836	2.9863	1090	0.4646	20.6950

Framework versions

• PEFT 0.17.1
• Transformers 4.57.6
• Pytorch 2.8.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2